Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers

Page created by Leroy Gibson
 
CONTINUE READING
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
ORIGINAL RESEARCH
                                                                                                                                            published: 30 March 2022
                                                                                                                                      doi: 10.3389/frsen.2022.860282

                                          Applying a Multi-Method Framework
                                          to Analyze the Multispectral Acoustic
                                          Response of the Seafloor
                                          Pedro S. Menandro 1*, Alex C. Bastos 1, Benjamin Misiuk 2 and Craig J. Brown 2
                                          1
                                              Marine Geosciences Lab, Departamento de Oceanografia e Ecologia, Universidade Federal do Espírito Santo, Vitória, Brazil,
                                          2
                                              Seascape Ecology and Mapping Lab, Department of Oceanography, Dalhousie University, Halifax, NS, Canada

                                          Improvements to acoustic seafloor mapping systems have motivated novel marine
                                          geological and benthic biological research. Multibeam echosounders (MBES) have
                                          become a mainstream tool for acoustic remote sensing of the seabed. Recently,
                                          “multispectral” MBES backscatter, which is acquired at multiple operating frequencies,
                                          has been developed to characterize the seabed in greater detail, yet methods for the use of
                                          these data are still being explored. Here, we evaluate the potential for seabed
                                          discrimination using multispectral backscatter data within a multi-method framework.
                                          We present a novel MBES dataset acquired using four operating frequencies (170, 280,
                                          400, and 700 kHz) near the Doce River mouth, situated on the eastern Brazilian continental
                          Edited by:      shelf. Image-based and angular range analysis methods were applied to characterize the
      DelWayne Roger Bohnenstiehl,        multifrequency response of the seabed. The large amount of information resulting from
      North Carolina State University,
                       United States
                                          these methods complicates a manual seabed segmentation solution. The data were
                       Reviewed by:
                                          therefore summarized using a combination of dimensionality reduction and density-based
                        Luis A. Conti,    clustering, enabling hierarchical spatial classification of the seabed with sparse ground-
      University of São Paulo, Brazil     truth. This approach provided an effective solution to synthesizing these data spatially to
              Bishwajit Chakraborty,
   Council of Scientific and Industrial    identify two distinct acoustic seabed classes, with four subclasses within one of the
              Research (CSIR), India      broader classes, which corresponded closely with seafloor sediment samples collected at
               *Correspondence:           the site. The multispectral backscatter data also provided information in likely, unknown,
               Pedro S. Menandro
         pedromenandro@gmail.com
                                          sub-surface substrate differences at this site. The study demonstrates that the adoption of
                                          a multi-method framework combining image-based and angular range analysis methods
                   Specialty section:     with multispectral MBES data can offer significant advantages for seafloor characterization
        This article was submitted to
                                          and mapping.
          Acoustic Remote Sensing,
              a section of the journal
                                          Keywords: seabed classification, multispectral backscatter, density-based clustering, seafloor mapping, benthic
        Frontiers in Remote Sensing
                                          habitat mapping, multibeam echosounder
         Received: 22 January 2022
          Accepted: 14 March 2022
         Published: 30 March 2022
                                          1 INTRODUCTION
                            Citation:
  Menandro PS, Bastos AC, Misiuk B        Technological innovations have advanced many areas of marine science over the past several
and Brown CJ (2022) Applying a Multi-
                                          decades, including seafloor mapping. Improvements to acoustic seafloor mapping systems have
   Method Framework to Analyze the
  Multispectral Acoustic Response of
                                          motivated novel benthic geological and biological research. Recently, seabed mapping science has led
                         the Seafloor.     to the development of a number of seabed classification tools (Fonseca et al., 2009; Rzhanov et al.,
      Front. Remote Sens. 3:860282.       2012; Lecours et al., 2016; Ierodiaconou et al., 2018; Masetti et al., 2018), which facilitate
     doi: 10.3389/frsen.2022.860282       multidisciplinary approaches to environmental management and conservation (Cogan et al.,

Frontiers in Remote Sensing | www.frontiersin.org                                  1                                          March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                               Multispectral Seafloor Acoustic Response

2009; Lee et al., 2015; Kirkman et al., 2019) across a range of              and presented the first results of multispectral MBES operating in
different scientific disciplines (Heap et al., 2014; Mosca et al.,            this manner. These studies reported improved seafloor
2016; Brooke et al., 2017).                                                  discrimination as a result of the multifrequency dataset, which
    Over the last 2 decades, multibeam echosounders (MBES)                   provides increased information enabling detailed comparison of
have become a mainstream tool for acoustic remote sensing of                 acoustic signatures between different seabed types.
the seabed (Brown et al., 2011; Menandro and Bastos, 2020).                      Despite these apparent advantages, the use of multifrequency
MBES are now widely used for hydrographic purposes and                       MBES for seafloor characterization is not well developed in many
applied research on the continental shelf (Innangi et al., 2015;             contexts. For example, single frequency backscatter datasets from
Rocha et al., 2020) and deep sea (Sen et al., 2016; Picard et al.,           both sidescan and MBES systems are useful as exploratory survey
2018; Stewart and Jamieson, 2019). In addition to measuring                  tools to obtain baseline knowledge of the composition of the
bathymetry, acoustic backscatter can be recorded from the MBES               seafloor (Augustin et al., 1996; Greene et al., 2013; Lucieer et al.,
signal. The concept of backscatter is basically the acoustic energy          2018). Because of the increased information they provide,
that returns from the seabed to the receiver of the sonar (Lurton,           multifrequency systems may be well suited to exploratory
2010), while it is still very complex structurally, following                tasks, yet the use of multiple frequencies also implies an
absorption, reflection, and scattering of the signal by the                   increase in dimensionality that must be resolved to realize the
seabed and water column, and influence of physical properties                 full benefit of these data. Additionally, several techniques may be
such as the acoustic frequency and the angle of incidence.                   used to derive secondary backscatter features for seafloor
Backscatter strength measurements have proven highly useful                  characterization, for example, by analyzing the acoustic
for discriminating benthic habitats (e.g., Costa, 2019; Harris and           angular response of the seafloor or the texture of the
Baker, 2020; Trzcinska et al., 2020) and surficial geology                    backscatter image (e.g., Blondel and Sichi, 2009; Alevizos and
(McGonigle and Collier, 2014; Montereale-Gavazzi et al.,                     Greinert, 2018; Fakiris et al., 2019). While highly useful for
2018). Most MBES backscatter seabed mapping analyses can                     classification, these approaches also serve to increase the
be divided into two domains: 1) image-based analyses,                        dimensionality of the data, and this effect is multiplicative in a
whereby backscatter data are compensated for angular                         multifrequency context. Furthermore, it is not immediately clear
dependence and analyzed as a single raster image (e.g.,                      which of these approaches or which information should be
Diesing et al., 2016; Ierodiaconou et al., 2018; Runya et al.,               prioritized for exploratory purposes, especially where there is
2021) and 2) angular range analysis, in which the full angular               little a priori knowledge on the composition of the seafloor or
response of backscatter is retained (Fonseca and Mayer, 2007;                sparse ground truth coverage.
Fonseca et al., 2009; Haris et al., 2011; Lurton et al., 2018). Angle-           Building on previous work, here we evaluate the potential for
varying gain (AVG), which is used to produce the mosaic for                  seabed discrimination using multispectral backscatter data within
image-based approaches, generally results in a loss of angular               a multi-method exploratory framework. We first present a novel
resolution, meaning that information may be lost that is specific             MBES dataset acquired using four operating frequencies (170,
to individual angles of incidence at the seafloor. Angular range              280, 400, and 700 kHz), with ground truth sediment information.
analysis, on the other hand, may be used to preserve angular                 Adopting a geoacoustic approach, image-based and angular range
detail at the cost of spatial resolution. Fonseca et al. (2009) point        analysis methods are applied to characterize the multifrequency
out that angular response and backscatter mosaic analysis can be             response of the seabed. The large amount of information resulting
considered as complementary methods, and the improvement of                  from these methods is then summarized using dimensionality
benthic maps through combined approaches has been                            reduction and density-based clustering, enabling hierarchical
demonstrated (Che Hasan et al., 2012; Che Hasan et al., 2014).               spatial classification of the seabed with sparse ground-truth.
    Currently, most MBES operate at a single acoustic frequency,
despite acknowledgement that the backscatter response of the
seafloor is frequency dependent. The difference between single-               2 MATERIALS AND METHODS
and multi-frequency backscatter analysis is comparable to mono-
and multi-frequency optical remote sensing (Diesing et al., 2016).           2.1 Data Acquisition and Processing
The field of terrestrial remote sensing is well developed with a              The current study focuses on a seafloor area (37 km2) adjacent to
fairly long history, and advanced models for image classification             the Doce River mouth, situated on the eastern Brazilian
using optical data employ a wide range of electromagnetic                    continental shelf, in the state of Espírito Santo. The Doce
frequencies simultaneously to achieve a more robust                          River is one of the main river deltas recognized along the
description of the land surface (Lu and Weng, 2007). Multi-                  Brazilian eastern coast (Dominguez, 2006), and is a major
frequency (aka, “multispectral”) acoustic remote sensing of the              sediment source to the adjacent continental shelf (Bastos et al.,
seafloor, in contrast, it is not yet as developed—though the                  2015). In terms of sedimentary processes, Quaresma et al. (2015)
potential advantages of multispectral acoustic data have been                characterized a deltaic lobe extending to 30 m depth, with an
recognized (Huges Clarke, 2015; Tamsett et al., 2016; Feldens                accumulation of fine sediments (>75% mud) in a main
et al., 2018). The recent implementation of multispectral MBES               depocenter south from the river mouth. Towards offshore
allows for bathymetry and backscatter acquisition at multiple                (depths >30 m), sandy facies are predominant (Figure 1)—a
frequencies simultaneously. Buscombe and Grams (2018), Gaida                 trend also observed and corroborated by Vieira et al. (2019).
et al. (2018), Brown et al. (2019), and Costa (2019) investigated            Though useful, previous research in this area has been conducted

Frontiers in Remote Sensing | www.frontiersin.org                        2                                     March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                                              Multispectral Seafloor Acoustic Response

 FIGURE 1 | Map of the study area (bathymetry adapted from Bastos et al., 2015), with sand content estimated by Quaresma et al. (2015). On the right-hand side,
 an illustration of multispectral MBES acquisition indicates frequency changes between sequential pings, and the nadir beam footprint sizes for each frequency.

at a coarse spatial resolution, and fine scale geomorphological and                    for each frequency was the only radiometric correction applied
substrate characterization are still lacking. The MBES and                            during the data acquisition—all other corrections (e.g., beam
sediment sample data sets presented here were analyzed to fill                         width) were applied during post processing. All systems were
this knowledge gap.                                                                   integrated using QPS QINSy 8.18.3 for data acquisition, and the
    A multispectral MBES dataset was acquired to provide high                         survey lines were planned with ~30% overlap. The dataset was
resolution information on the geomorphology and substrate                             assessed during the survey to ensure data quality using QPS
distribution of the study area. Bathymetry and backscatter                            Qimera 2.0, and post-processing was carried out using QPS
were acquired using an R2Sonic 2024 echosounder, with the                             Qimera and QPS FMGT 7.9.5 (Fledermaus Geocoder Toolbox,
sonar head deployed through a moon pool in the steel-hulled                           hereafter FMGT).
survey vessel, “Santa Edwiges” (19.6 m in length). The MBES was                          In compliance with the International Hydrographic
configured to collect data by sequentially pinging at 170, 280, 400,                   Organization, bathymetric processing consisted of manually
and 700 kHz (true frequency of 697.674 kHz) operating                                 cleaning erroneous soundings and tidal correction, which was
frequencies in equiangular mode. The 700 kHz frequency did                            performed using data extracted from an ADCP moored adjacent
not operate during the entire survey due to the effects of depth                      to the study area. The dataset was filtered for each operating
and water turbidity on data quality. Specifications such as power,                     frequency in QPS Qimera. Digital bathymetric models (DBM) for
pulse length, gain, and spreading were held constant. A 90°                           each frequency were exported at a 2 m horizontal resolution.
angular sector and 256 beams were used for all frequencies                               Although there is still no unified standard for processing
except 700 kHz, which is limited to a 70° sector. Because the                         backscatter, the multispectral data were processed following
receiver array is flat, the size of the receive beam grows with                        recommendations in Lamarche and Lurton (2018) and
distance from nadir (i.e., the lower the grazing angle, the larger                    Schimel et al. (2018). The main steps involve the frequency
the beam footprint; Figure 1). The transmit array is round and                        filtering, correction of acquisition parameters for each
does not suffer the same issue.                                                       frequency (gain, transmit power, pulse length, beam width)
    The MBES system was paired with a POS MV Wave Master                              and AVG correction. Backscatter mosaics were exported at
Inertial Navigation System (INS), with differential positioning.                      1 m resolution. All backscatter processing was completed
During the survey, SVP casts were deployed every 3 hours using a                      using FMGT.
Valeport Mini configured to collect sound velocity, salinity,
temperature, and pressure. These data are essential to ensure                         2.2 Multispectral Backscatter Analysis
the correction of sound velocity effects and to calculate and apply                   Elements of both image-based and angular range approaches
the absorption coefficient. Application of absorption coefficients                      were used to analyze backscatter data and explore the

Frontiers in Remote Sensing | www.frontiersin.org                                 3                                           March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                               Multispectral Seafloor Acoustic Response

 FIGURE 2 | Flow indicating the steps of the methodological framework.

multispectral response of the seabed. The methodological                     Fonseca et al. (2009) suggested that the angular response curve
framework of the analysis is summarized in Figure 2,                         can be sectorized according to the angular range—comprising, for
outlining each approach and its expected results.                            example, the near range region (near nadir) from 0 to 25°, the far
                                                                             range region from 25 to 55°, and the outer range region from 55 to
2.2.1 Image-Based Analysis                                                   85°. Angular response curves were extracted at each ground truth
The image-based analysis can be subdivided into three steps. The             sample site through swath profiles for comparison with
first step consists of visual interpretation of the mosaics and               geophysical models and comparison between frequencies.
assessment of the uncalibrated backscatter values for each                   Additionally, corrected and georeferenced soundings were
frequency using profiles, graphs, and descriptive statistics. This            extracted from FMGT outputs for each 10° sector from the
step is critical for obtaining baseline understanding of the data            nadir (i.e., 1°–10°, 11°–20°, 21°–30° and 31°–40°) using a custom
and is often combined with other analyses (Parnum and Gavrilov,              R script. These results were rasterized and integrated with the
2011). General data exploration and visual comparisons were                  layers from with other approaches within ArcGIS for further
performed using ArcGIS and R. In the second step, textural                   analysis.
parameters were extracted from the mosaics, which is a common
technique for classifying acoustic data (Diesing et al., 2016;               2.3 Density-Based Clustering
Diesing et al., 2020). Here, textural analyses were based on                 The large amount of information from the different analysis
grey-level co-occurrence matrices, using the GLCM v 1.6.5                    approaches (Multispectral Backscatter Analysis) was summarized
package available in R (Haralick et al., 1973). For each mosaic,             spatially using density-based clustering. Unlike many clustering
GLCM variance measurements were extracted, based on how                      methods that rely on measures of multivariate distance to identify
frequently different combinations of neighboring pixels occur (in            clusters, these approaches identify clusters using estimates of
a 5x5-pixel window). The third step was the generation of a                  multivariate data density (Hahsler et al., 2019; Hashler and
multispectral composite image (i.e., a false-colour RGB) using               Piekenbrock, 2021). Density-based clustering offers several
each frequency as a spectral band. This approach has not yet been            advantages compared to many other approaches, such as the
widely applied, likely as a function of the recent development of            automatic identification of cluster numbers, rejection of outliers
multispectral MBES, but has great potential to improve habitat               that do not fall within any dense regions, and clustering of data of
discrimination in combination with other terrain attributes.                 arbitrary multivariate shape (Ester et al., 1996; Kriegel et al.,
                                                                             2011). Ordering Points to Identify Clustering Structure (OPTICS)
2.2.2 Angular Range Analysis                                                 is a method that can additionally identify clusters of varying
The second approach applied to the multifrequency corrected                  density (Ankerst et al., 1999). OPTICS is an extension of the
backscatter dataset was angular range analysis (ARA). The                    DBSCAN algorithm (Ester et al., 1996), which facilitates data
backscatter angular dependence (the acoustic response across                 exploration through “reachability plots”—one-dimensional
the range of ensonification grazing angles), provides information             representations of the “reachability distance” between points
that can be used for improved seabed classification (Che Hasan                (Hahsler et al., 2019). These can be used to inform the
et al., 2014; McGonigle and Collier, 2014; Fezzani and Berger,               clustering outcome, and enable additional flexible solutions
2018), or the inference of seafloor physical properties such as               such as hierarchical density-based clustering. Despite the
grain size and acoustic impedance (Fonseca and Mayer, 2007).                 widespread application of unsupervised approaches for seabed

Frontiers in Remote Sensing | www.frontiersin.org                        4                                     March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                                                 Multispectral Seafloor Acoustic Response

 FIGURE 3 | Workflow for density-based clustering of the highly dimensional multi-method backscatter analysis. Illustrative scheme of density-based clustering was
 adapted from Hahsler et al. (2019).

mapping and classification, and the potential benefits offered by                         using the parameters (the density search radius) and (a threshold
density-based clustering, it has received relatively little attention                   number of points used for reachability calculations) to visualize the
in this field (but see Le Quilleuc et al., 2021).                                        data density and estimate the number of clusters. The reachability
    Results from both image-based and angular range approaches                          plot suggested a hierarchical data structure, and two sets of clusters
were used for clustering, with the exception of the frequency of                        were extracted at coarse and fine levels of detail. These indicated
700 kHz, which has a smaller coverage area and was not                                  broad data groupings, which contained smaller sub-clusters at a
considered for this analysis. These layers included AVG                                 higher level of detail. The clustering results were exported to GIS
backscatter mosaics for each frequency (170, 280, and                                   format to generate benthic map predictions.
400 kHz), GLCM variance for each frequency (170, 280, and
400 kHz), and backscatter values from four different angular                            2.4 Ground Truth Information
sectors (1–10°, 11–20°, 21–30°, 31–40°) for each frequency                              After preliminary analysis of the backscatter mosaics, samples of
(170, 280 and 400 kHz), totaling 18 rasters (3 AVG mosaics, 3                           surficial sediments were collected at 12 sites within the survey
rasters from GLCM results, and 12 rasters from ARA). We                                 area (Figure 1). Sediments were collected using a Van Veen grab
applied the methods presented in Che Hasan et al. (2012) in                             sampler in July 2021 (nearly 2 years after the acoustic data
order to aggregate the acoustic data to object-based image                              survey). To account for the temporal offset between the
segments. The three-band RGB backscatter mosaic was                                     collection of sediment samples and acoustic data, we verified
segmented using the “mean shift” algorithm (Yizong, 1995;                               the differences and the general trend from the sediment
Comaniciu and Meer, 2002) in ArcGIS Pro using the                                       distribution (Quaresma et al., 2015; Vieira et al., 2019). The
maximum spatial and spectral detail, but a minimum segment                              grain size analysis provides results as percentage gravel, coarse
size of 2000 pixels. The mean shift is a “mode-seeking” algorithm                       sand, medium sand, fine sand, silt and clay, which were used to
that assigns data points to clusters based on their position relative                   ground-truth results from the multiple backscatter analysis
to the local maxima of a multivariate kernel density estimate. The                      approaches and density-based clustering.
values of the 18 rasters were stacked and averaged over the
resulting segments, which allowed for continuous estimates of
the angular backscatter values across the full extent of the study                      3 RESULTS
area, and a reduction of data volume by several orders of
magnitude, producing 6902 segments (step 2 of Figure 3). The                            3.1 Multispectral Bathymetry
values for each segment (18 acoustic variables) were converted to                       The DBM outputs, which contain all processed depths from all
matrices in R, and principal component analysis (PCA) was                               frequencies combined, ranged from 10 to 32 m, and revealed a
applied to eliminate collinearity and reduce dimensionally.                             predominantly flat bottom with locally steep features. These were
The first two components explained 95.4% of the total                                    evident in maps of local slope and roughness, and ridges oblique
variance and were retained for clustering.                                              to the coast were visible at the deep end of the profile (Figure 4).
    Principal components of the image-based segments were                                   The spatial distribution of depth differences was also
clustered using OPTICS to identify acoustically distinct seabed                         investigated. DBM surfaces generated for each operating
types across the study area. OPTICS was implemented using the                           frequency were subtracted from each other using raster
DBSCAN package in R (Figure 3; Ester et al., 1996, Campello et al.,                     algebra to visualize the distribution of depth differences
2013; Hahsler et al., 2019). The reachability distance was first plotted                 across the study area (Figure 5). In general, regions close to

Frontiers in Remote Sensing | www.frontiersin.org                                   5                                            March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                                          Multispectral Seafloor Acoustic Response

 FIGURE 4 | (A) Digital bathymetric model; (B) bathymetric profile; (C) slope; (D) ruggedness.

 FIGURE 5 | Bathymetric difference surfaces (in meters) computed from the subtraction between the digital bathymetric models of each frequency.

Frontiers in Remote Sensing | www.frontiersin.org                                6                                        March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                           Multispectral Seafloor Acoustic Response

higher slope values (steep feature and the edges of oblique               backscatter feature running south-north through the survey area
ridges) show higher bathymetric difference between                        all frequencies show high backscatter values, and differences are
frequencies. Differences between 170 and 280 kHz were, on                 not apparent (Figures 6, 7). The 700 kHz mosaic appeared
average, 0.1 m greater than the differences between 280 and               oversaturated. Figure 7 suggests a general trend of increasing
400 kHz, and differences between 170 and 400 kHz were                     backscatter and greater agreement between frequencies with
greater still. The differences between 700 and 170 kHz were               increasing distance from northwest to southeast across the
the greatest observed, whereas differences between 400 and                survey area.
700 kHz were three times lower. The bathymetric differences                   Uncalibrated backscatter values for the frequencies of 170, 280
(which can reach up to 0.2 m) observed between the                        and 400 kHz were plotted as violin plots for the whole area
frequencies largely conformed to expectation given the                    (Figure 8A) and also for all frequencies using only the 700 kHz
current understanding of substrate-frequency response and                 coverage area (Figure 8B). These plots present the kernel
the bottom type in the study area.                                        probability density of the data. Considering the entire area, the
                                                                          lowest median backscatter values (−30.63 dB) were observed
3.2 Multispectral Backscatter                                             using the 400 kHz frequency. The 170 kHz backscatter median
3.2.1 Image-Based Analysis                                                was greater (−24.97 dB), with the smallest interquartile range of
3.2.1.1 Assessment of Mosaics and Uncalibrated Backscatter                values (between Q1 and Q3), which is also reflected in the
Values                                                                    concentration of data observed in the kernel density curve
Figure 6 presents the uncalibrated backscatter mosaics for each           with leptokurtic tendency and positive asymmetry (see
frequency at 1 m resolution using the same range of values for the        Figure 8B). The 280 kHz data presented a higher median
greyscale (−10 and −38 dB). Overall, the shallower region presents        value (−26.54 dB) than the 400 kHz, and both frequencies
the lower backscatter values, and the oblique elongated bathymetric       presented similar data distributions (Figure 8A), recognizing
features correspond with higher backscatter values.                       two distinct sets of backscatter values (the main one having a
   From visual analysis, differences between the frequencies              modal value close of −27.80 dB for 280 kHz and −31.89 dB for
appear greater in low backscatter regions, while over the high            400 kHz). The better recognition of two main groups of

 FIGURE 6 | Backscatter mosaics for each operating frequency.

Frontiers in Remote Sensing | www.frontiersin.org                     7                                    March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                                                 Multispectral Seafloor Acoustic Response

 FIGURE 7 | Top: comparison of differences in multispectral backscatter intensities along the transect shown in the mosaic (top left). Bottom: comparison of
 differences in multispectral backscatter intensities along the transect shown in the mosaic inset (bottom left).

 FIGURE 8 | (A) Violin plots presenting backscatter data (170, 280, and 400 kHz) for the whole area, and (B) for all frequencies, spatially restricted within the
 700 kHz coverage area.

Frontiers in Remote Sensing | www.frontiersin.org                                   8                                            March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                                                Multispectral Seafloor Acoustic Response

 FIGURE 9 | Composite image (RGB) representing the three frequencies (A) For the whole study area; (B) Frame over the sandy feature. Pixels are stretched to the
 minimum and maximum values for each band. No changes of color rendering parameters were applied (hue, saturation and contrast).

 FIGURE 10 | GLCM variance results for each frequency.

Frontiers in Remote Sensing | www.frontiersin.org                                  9                                            March 2022 | Volume 3 | Article 860282
Applying a Multi-Method Framework to Analyze the Multispectral Acoustic Response of the Seafloor - Frontiers
Menandro et al.                                                                                                         Multispectral Seafloor Acoustic Response

 FIGURE 11 | Angular range analysis results obtained by extraction of backscatter values for each 10° angular sector.

 FIGURE 12 | Angular dependence curves for each frequency extracted in each sample site.

Frontiers in Remote Sensing | www.frontiersin.org                               10                                       March 2022 | Volume 3 | Article 860282
Menandro et al.                                                                                                                        Multispectral Seafloor Acoustic Response

backscatter values by the higher frequencies can also be observed                            indicate low backscatter values for all frequencies. The greatest
by the standard deviation of the dataset for each frequency                                  differences between the bands are indicated by red shades, which
(170 kHz: 3.35 dB, 280 kHz: 4.64 dB, 400 kHz: 5.09 dB).                                      indicates higher values for the 170 kHz signal, and in light blue
   Over the 700 kHz coverage (Figure 8B), the lowest median                                  areas, which indicate higher values for the combination of the two
backscatter values (−33 dB) were observed for the 400 kHz data.                              higher frequencies (280 and 400 kHz).
The similar shape of the density curves (Figure 8B) is expected, once
the seabed appears to be more homogeneous in this area in                                    3.2.1.3 Textural Analysis (GLCM)
accordance with visual analysis of the mosaics. Considering the                              The general trend of textural variations can be visualized using
oversaturated mosaic, and that little is currently known concerning                          the GLCM variance (Figure 10). Higher values occurred over the
the acoustic response of 700 kHz frequency, we reserve a detailed                            high backscatter feature, especially at 280 kHz. Lower values
image-based interpretation of the 700 kHz data. Further                                      occurred in the low backscatter region, mainly at 400 kHz.
investigation would be useful to inform on the effects of a turbid
water column on such high-frequency MBES soundings.                                          3.2.2 Angular Range Analysis
                                                                                             The angular range analysis provides results that can be explored
3.2.1.2 Multispectral Composite Image                                                        using geophysical and empirical approaches, including physical
The backscatter mosaics were combined to produce a three-band                                model comparison and empirical parameters to distinguish
composite RGB image (Figure 9). In the RGB mosaic, it is                                     seafloor types, and also angular dependence curves of different
possible to recognize zones where all frequencies show                                       substrates (Lurton and Lamarche, 2015). The angular backscatter
similarly high backscatter values (white regions), which                                     values were interpolated spatially using Inverse Distance
coincide with the elongated and oblique features. Darker zones                               Weighting (IDW) in ArcGIS Pro to produce grids of 1 × 1 m

 FIGURE 13 | Reachability plot of a hierarchical clustering with (A) two main clusters (reachability distance = 0.88) and noise; and (B) clusters at detailed hierarchical
 level (reachability distance = 0.475).

Frontiers in Remote Sensing | www.frontiersin.org                                       11                                              March 2022 | Volume 3 | Article 860282
Menandro et al.                                                                                                                           Multispectral Seafloor Acoustic Response

cells (Figure 11). The results demonstrated that, for most of the                              at 400 kHz—but high values were still observed at the elongated
area, decreasing backscatter was associated with increased                                     feature.
incidence angle and frequency. The exception occurs over the                                       Angular response curves were extracted for all frequencies to
high-backscatter feature, which marks a region with high                                       characterize the acoustic response of the seabed at each sample
scattering across the entire swath.                                                            site (Figure 12). The angular response curves can be broadly
    The 1–10° sector shows high backscatter values throughout the                              described in three main groups. The curves at sample sites Doce
area and similar results for all frequencies. Compared to results                              01 and Doce 02 show similar shapes for all frequencies, with
presented for the lower frequencies at 11–20° sector, the 400 kHz                              decreasing backscatter values as the frequency increases (with
frequency seems to achieve a better power of seafloor                                           exception of 700 kHz). The shape shows low backscatter level loss
discrimination for the shallower region. Three distinct regions                                for the first 20° from the nadir, followed by a greater decrease in
are clearly visible using the 21–30° sector. Additionally, the                                 backscatter with increasing incidence angle. The curves extracted
400 kHz data within this sector may suggest that the shallow                                   at Doce 07, 09, 11, and 12 are comparatively flat, with high
region is composed of different material than the elongated                                    backscatter values and very slight differences between
feature, even though they appear similar at other frequencies                                  frequencies. The third group is characterized by a high
and angular sectors. Lower backscatter was observed using the                                  backscatter level loss for all frequencies from nadir to the
outermost angular sector over much of the study area—primarily                                 outer beams, and lower values at 400 kHz.

 FIGURE 14 | (A) Density-based clusters predicting for the study area showing two main clusters; (B) clusters at detailed hierarchical level; (C) grain size results on top of
 clustering solution; (D) results of each grain size fraction for each sample; (E) sediment sample at Doce 07 sample site; (F) sediment sample at Doce 10 sample site.

Frontiers in Remote Sensing | www.frontiersin.org                                         12                                               March 2022 | Volume 3 | Article 860282
Menandro et al.                                                                                                             Multispectral Seafloor Acoustic Response

TABLE 1 | Characteristics of each cluster.

                  Ground truth samples                 Angular response curve                                         Sediment properties

Cluster 1         Doce   04, 05, 08 and 10   High backscatter level loss                      Mostly mud or fine sand
Cluster 1a        —                          —                                                —
Cluster 1b        Doce   06                  High backscatter loss                            Predominantly fine sand, with more than 10% mud
Cluster 1c        —                          —                                                —
Cluster 1d        Doce   01 and 02           Increased backscatter loss past 20°              More than 95% fine sand
Cluster 2         Doce   09, 11 and 12       Low backscatter loss                             Predominantly medium sand, with less than 10% gravel
Noise/Outlier     Doce   03 and 07           High (Doce 03) and low (Doce 07) BS level loss   Fine sand with less than 30% mud (Doce 03) and gravelly sand (Doce 07)

3.3 Density-Based Clustering and Mapping                                                Cluster 1c also does not have sediment samples, but it can
The combination of PCA and density-based clustering enabled                             be interpreted as similar to cluster 1b, according to the angular
synthesis of a large number of backscatter data layers to inform                        sector maps (Figure 11). Cluster 1d has two samples and can
seabed classification. PCA reduced the 18 backscatter-derived layers                     be defined as fine sand. Some areas were left unclassified but
to two principal components that explained the majority of variance                     may comprise two additional substrate types—one with higher
(>95%) in the dataset, suggesting substantial collinearity among the                    coarse sand/gravel contents (sample Doce 07) and the other
variables. The OPTICS reachability plot from these principal                            with extremely low backscatter values, indicating a muddy
components (Figure 13) enabled identification of clusters at two                         bottom. It is possible that the latter has undergone
hierarchical levels. Low reachability distances, represented as valleys                 modifications during the time between the collection of
in the plot, indicate dense data regions, which are separated by                        acoustic data and ground truthing.
“peaks” representing regions of low data density. First, two broad,
well-defined clusters were evident at a reachability distance of 0.88
(Figure 13A). These results agree with general trends observed for                      4 DISCUSSION
many of the backscatter layers (Figures 6, 10, 11), in which the study
area is separable into distinct areas of low and high backscatter                       Backscatter data has been analyzed in a variety of ways for
intensity and variance. Cluster 1, which occurred across the full                       seafloor classification, including interpretation of backscatter
extent of the study area (Figure 14A), could be further divided at a                    mosaics, textural analysis (Runya et al., 2021), image-based
reachability distance of 0.48 to produce four sub-clusters. Unlike the                  analysis (Ierodiaconou et al., 2018), and techniques involving
broader Cluster 1, the four sub-clusters show a distinct gradient from                  angular range analysis (Fonseca and Mayer, 2007). Advantages
the coast (Figure 14B), which was not immediately apparent from                         and disadvantages of each make the choice of methodology
visual analysis of the various backscatter layers. Outliers appear in                   challenging, and their combined use may achieve better results
regions of extreme high and low backscatter values.                                     by uniting the spatial resolution of the mosaics with the
                                                                                        enhanced information supplied from angular analysis
3.4 Grain Size                                                                          (Fonseca et al., 2009; Che Hasan et al., 2014). Several
The results from grain size analysis (Figures 14C,D) show that                          supervised and unsupervised techniques have been applied
samples Doce 01 and Doce 02 are composed primarily of fine                               in seabed classification (Stephens and Diesing, 2014; Diesing
sand. Doce 03 and Doce 04 share a similar mud content and both                          et al., 2020) and incorporated in segmentation algorithms,
stations are close to bathymetric anomalies (high slope and high                        including different clustering approaches (Le Bas, 2016;
ruggedness sites). Samples Doce 05, Doce 08 and Doce 10 (Figure                         Masetti et al., 2018). Density-based clustering has received
14F) had an increased mud content, while sample sites at the high                       little attention for seabed classification, and was successfully
backscatter features (Doce 07, 09, 11, 12) - e.g., Figure 14E - had                     applied here to synthesize different analysis approaches to
higher proportions of medium and coarse sand fractions, with                            produce a classified output.
low amounts of mud.
   Although there not enough samples to estimate the                                    4.1 Benthic Map Prediction
classification accuracy, the classes seem to distinguish benthic                         Highly dimensional input data resulting from multiple analysis
clusters through the multivariate combination of the results.                           approaches can be difficult to synthesize spatially, especially given
Table 1 summarizes the main properties for each cluster defined.                         sparse ground truth information, yet density-based clustering
   Cluster 1 contained the greatest number of sediment                                  provides an objective means to identify well-defined seabed units
samples and can be characterized as muddy fine sand.                                     under such circumstances. Here, backscatter mosaics (Figure 6)
Cluster 2 is medium sand, with a smaller coarse sand                                    and the composite RGB (Figure 9) potentially suggest two or
fraction. Cluster 1a contains no ground truthing validation,                            three seabed classes, while the angular response curves at the 12
but seems to indicate a transition between clusters 1 and 2.                            sample sites (Figure 12) suggest at least three classes. GLCM
Cluster 1b seems to have characteristics close to cluster 1a,                           products and spatial analysis of different angular sectors further
likely with higher mud content, or morphological                                        confound a unified manual seabed segmentation solution. The
characteristics that provide a soft backscatter response.                               combination of image-based segmentation, dimensionality

Frontiers in Remote Sensing | www.frontiersin.org                                  13                                        March 2022 | Volume 3 | Article 860282
Menandro et al.                                                                                              Multispectral Seafloor Acoustic Response

reduction, and density-based clustering provide an objective                 could be further investigated by observing the influence of survey
solution to synthesizing these data spatially to identify distinct           azimuth (Lurton et al., 2018), presence of bedforms, the variety of
seabed types. Similar solutions have applied PCA for                         microtopography, and even more ground truthing to check
dimensionality reduction of backscatter data followed by                     lateral variability.
k-means clustering for unsupervised classification (e.g.,                         The use of multispectral technology offered increased
Alevizos et al., 2018), and other approaches towards similar                 seafloor discriminatory power based on the different
ends have included combinations of self-organizing maps and                  frequency response of the seafloor across the frequency
fuzzy clustering (Chakraborty et al., 2015), and Bayesian                    range, achieving a benthic prediction in agreement with
probability estimation (Amiri-Simkooei et al., 2009; Simons                  earlier studies in this region (Quaresma et al., 2015; Vieira
and Snellen, 2009). Several properties of density-based                      et al., 2019), and making it possible to improve the seafloor
clustering are desirable in an exploratory context though,                   classification. Quaresma et al. (2020) found lower seabed
including automatic identification of number of clusters,                     density values in the shallow region close to the mud
enhanced data visualization and hierarchical clustering                      depocenter (Figure 1), and high density values (>1400 kg/
solutions, and rejection of outliers that are dissimilar from                m³) in regions with depths greater than 30 m. Here, the
other well-defined clusters. These qualities are lacking in many              lower uncalibrated backscatter values appear close to the
other common approaches, which may be well-suited to                         steep feature (i.e., high slope–see Figure 4) and on the edge
classifying gradational or poorly defined seabed types (e.g.,                 of the sandy elongated feature, where the bathymetric
k-means, ISO, agglomerative clustering).                                     differences (Figure 5) between the frequencies reached up to
    Quantifying the advantages of using multispectral backscatter            10 cm, indicating a mud accumulation partially driven by the
data as a predictor for different seafloor types is beyond the scope          seafloor morphology. In these regions, the lower frequency was
of the current study, but the findings suggested that the                     presumably detecting the sub-surface sediment characteristics,
multifrequency acoustic data provided greater discrimination                 while the higher frequency reflected primarily off the surface
of muddy and fine sand sediments than coarser sediments,                      sediments of the seafloor, potentially indicating a thick, muddy
aligned with findings from other studies (Gaida et al., 2018;                 deposit in this area. Although results from the bathymetric data
Brown et al., 2019; Costa, 2019). Part of the data identified as              (e.g., slope, ruggedness, bathymetric differences) played a
noise by the OPTICS algorithm is located in the region of highest            secondary role compared to the backscatter here, they can
backscatter, where the sediment shows the higher gravel and                  serve as important predictors for seabed classification
coarse sand content. Conversely, multiple distinct clusters were             (Walbridge et al., 2018; Lucatelli et al., 2019), and some
identifiable in muddy sediments. We believe this to be a property             derivatives may be more important than bathymetry itself
of the acoustic data, which were highly variable and                         (Trzcinska et al., 2020).
inhomogeneous at the high backscatter regions. Further                           The analysis of the composite band image proved to be an
investigations using multispectral backscatter for a greater                 interesting tool to visualize the results in an integrated way,
range of seabed types is necessary to determine the potential                making it possible to observe regions where all frequencies
of this technology for substrate discrimination. This is also the            indicated similar characteristics for distinct seabed types (for
first application of density-based clustering methods in this                 example, darker tones indicating more muddy bed, and whiter
context as far as we know, and further comparison with other                 tones indicating more sandy sediment). Nevertheless, the
clustering methods in various environments would be of great                 observation of the RGB mosaic also allowed to recognize
interest.                                                                    differences in acoustic responses in some regions, such as in
                                                                             the redder region near to the sample sites Doce 01 and Doce
                                                                             02–indicating higher backscatter values for the lower frequency,
4.2 Exploring Multispectral Backscatter                                      and a lighter bluish region near to sample site Doce 11)–coloring
Response Through Image-Based and                                             that indicates higher backscatter values for the combination of the
Angular Range Analysis Approaches                                            higher frequency bands (280 and 400 kHz). Quantifying the gain
Despite limitations associated with the use of uncalibrated                  due the use of such methodology is beyond the scope of this work,
backscatter data (Lurton and Lamarche, 2015; Schimel et al.,                 but the result brings new information about the spectral signature
2018; Malik et al., 2019), the results provided basic information            of certain types of bottoms, and helps to consolidate the potential
for interpretation of different seafloor types, and provided useful           to improve the distinction and classification of the seafloor using
information on the acoustic response of the seabed. For most                 multispectral backscatter (Costa, 2019).
areas, the median uncalibrated backscatter values from the                       Results from the GLCM analysis provide information that may
mosaics for all frequencies were low (slightly higher for lower              aid in interpreting geoacoustic properties of the seabed. Jackson
frequency), which is characteristic of soft muddy deposits (Brown            and Briggs (1992) found that the scattering interface may be a
et al., 2019; Diesing et al., 2020). It is also worth noting that the        dominant factor controlling backscatter in sandy bottoms, while
characteristics of sub-bottom deposits can influence the acoustic             volume homogeneity is increasingly important in bottoms of
response, especially for lower frequencies (Jackson and                      higher mud (and lower medium-coarse sand) content. Here, the
Richardson, 2007; Williams et al., 2009; Feldens et al., 2018).              highest GLCM variance values were observed at the oblique and
The acoustic response from sandy features (both for image-based              elongated sandy features, indicating higher uncalibrated
and ARA results) is largely influenced by interface scattering, but           backscatter values due to mosaic grey-level values—particularly

Frontiers in Remote Sensing | www.frontiersin.org                       14                                    March 2022 | Volume 3 | Article 860282
Menandro et al.                                                                                              Multispectral Seafloor Acoustic Response

at the 280 kHz operating frequency (Figure 10). Hughes Clarke               (2021) summarized backscatter values and angular response
(2015) also noted an increase in backscatter intensity over sand            curves extracted from different studies over a range of seabed
sheets with increasing frequency, yet retained uncertainty                  types. Multispectral backscatter facilitated this relationship with
regarding textural classification, as the bottom ruggedness                  the results presented in the angular response curves: higher
measurement depends on the acoustic wavelength. Closer to                   backscatter values were observed for higher frequencies in fine
the river mouth, differences in GLCM variance may be                        sand bottoms here, which appeared in a similar shape to the very
associated with other variables that influence the backscatter               fine sand in Trzcinska et al. (2021). Higher backscatter values for
response, such as the thickness and homogeneity of the                      280 kHz on a medium sand bottom here also appeared similar to
muddy subsurface deposit. In addition to informing                          their findings for sandy gravel (Trzcinska et al., 2021). For muddy
classification, the GLCM variance contributed to visual                      bottoms, this relationship is made difficult by the suspected
interpretations, providing results that assist with backscatter             presence of a heterogeneous subsurface. For a better
analysis and classification of the seabed (Samsudin and Hasan,               understanding of the results, it would certainly be interesting
2017; Runya et al., 2021).                                                  to include some subsurface analysis (e.g., sub bottom profiler or
    While the 30–60° angular range is ideal for many seabed                 even subsurface sediment samples) and observe the ARA results
mapping applications (Lucieer et al., 2018), the utility of                 at smaller angular intervals with more ground truthing.
angular sector may change with frequency, and therefore, the                   The seafloor variability within the swath is one of the
extraction of backscatter values for different angular sectors may          difficulties related to the angular range analysis, but the use of
provide increased information with which to inform seabed                   RGB image segmentation before clustering the variables of both
classification. The 1–10° sector contains the most specular                  approaches was an effective way to combine the methods and
reflection, and here, appeared to show low contrast between                  explore the dataset more adequately. The combination of such
different seabed types. The 11–20° data provided useful                     approaches applied to multispectral backscatter is still scarce in
information at the fine sandy region, primarily at the higher                the scientific literature, but showed potential to better explore the
frequencies. The angular response curves for this region show low           advantages of each method using multifrequency dataset, even
backscatter loss for the first 20°, followed by a higher backscatter         not involving bathymetric predictors.
loss with increasing incidence angle. The 21–30° angles allowed
for distinguishing bottom types using the lower frequencies. The
400 kHz data at these angles allows for visually distinguishing             5 CONCLUSION
differences in bottom type close to the river mouth, which were
not apparent at lower angles. The 31–40° sector showed high                 Density-based clustering enabled the integration of results from
contrast between the sandy feature and the rest of the region.              multiple approaches to produce a comprehensive unsupervised
While the distinction between fine sand and muddy regions was                classification of the seabed using multispectral acoustic data. Both
not always clear using image-based methods, the use of                      image-based and ARA approaches produced input components
backscatter at different angular sectors and frequencies                    for the final classification, and the combined use of both helped to
enabled their discrimination, further indicating the benefits of             ensure that all relevant information was included. While
combined analysis approaches for enhanced seafloor                           backscatter mosaics, the composite RGB and angular response
discrimination (Che Hasan et al., 2012; Ierodiaconou et al., 2018).         curves potentially suggest two or three seabed classes, the final
    The ARA results appeared to correspond to the composition of            clustering identified two distinct acoustic seabed classes, with four
sediment samples. Those with a large proportion of mud showed               subclasses within one of the broader classes, which corresponded
a greater rate of backscatter decrease with incidence angle than            closely with seafloor sediment samples collected at the site. In
samples describing coarser sediments (Figure 12), and the                   addition, angular range analysis suggested that information from
greatest differences between frequencies. This trend was most               different angular sectors may be informative for different seabed
apparent from the 400 kHz data, which is consistent with lower              types, and using different frequencies. Therefore, multispectral
uncalibrated backscatter values mapped by the image-based                   backscatter appeared to offer greater advantages in terms of
results and with the acoustic influence of the muddy deposit                 discrimination seafloor in mud and fine sand bottom types
homogeneity. Gaida et al. (2018) observed an increasingly flat               than in coarser sediments.
angular response curve at 400 kHz in soft sediments compared to
lower frequencies, which they attributed to an increased
sensitivity to roughness at higher frequencies. Here, the                   DATA AVAILABILITY STATEMENT
angular response curves observed for sandy sample sites (Doce
07, 09, 11, and 12) exhibited flatter shape, similar to acoustic             The raw data supporting the conclusion of this article will be
model curves (Applied Physics Laboratory, 1994). Combining                  made available by the authors, without undue reservation.
with the other results, it can be concluded here that the flatter the
angular response curve, the recognition of medium/coarse sandy
bottoms is better observed in the image textural analysis,                  AUTHOR CONTRIBUTIONS
corroborating the greater influence of the scattering interface.
Deriving consistent trends between backscatter and frequency                PM: field work, conceptualization, formal analysis, investigation,
over a range of seabed types remains difficult, yet Trzcinska et al.         writing–original draft; AB and CB: conceptualization, review and

Frontiers in Remote Sensing | www.frontiersin.org                      15                                     March 2022 | Volume 3 | Article 860282
Menandro et al.                                                                                                                        Multispectral Seafloor Acoustic Response

editing; BM: conceptualization, formal analysis, investigation,                             ACKNOWLEDGMENTS
review and editing.
                                                                                            PM was granted an international scholarship from the
                                                                                            Coordenação de Aperfeiçoamento de Pessoal de Nível
FUNDING                                                                                     Superior–Brazil (CAPES)–Finance Code 001 at Dalhousie
                                                                                            University to work in partnership in the Ocean Frontier
Data collection was carried out as part of Impactos Lama de Rejeito nos                     Institute Benthic Ecosystem Mapping and Engagement
Habitat Marinhos da Foz do Rio Doce (FAPES Grant 77683390) and                              (BEcoME) Project. AB is a Research Fellow of CNPq (PQ1D).
Environmental Monitoring Program (PMBA) (Fundação Renova-                                   We are grateful to all laboratory colleagues that collaborated
FEST-RRDM). This research is part of the Ocean Frontier Institute                           during field working or helping in data processing and further
Benthic Ecosystem Mapping and Engagement (BEcoME) Project.                                  discussions.

                                                                                               Map and Fuzzy Algorithm) Using Backscatter Data for Segmentation and Fine-
REFERENCES                                                                                     Scale Roughness Characterization of Seepage-Related Seafloor along the
                                                                                               Western Continental Margin of India. IEEE J. Oceanic Eng. 40, 3–14. doi:10.
Alevizos, E., and Greinert, J. (2018). The Hyper-Angular Cube Concept for                      1109/joe.2013.2294279
   Improving the Spatial and Acoustic Resolution of MBES Backscatter                        Che Hasan, R., Ierodiaconou, D., and Laurenson, L. (2012). Combining Angular
   Angular Response Analysis. Geosciences 8, 446. doi:10.3390/                                 Response Classification and Backscatter Imagery Segmentation for Benthic
   geosciences8120446                                                                          Biological Habitat Mapping. Estuarine, Coastal Shelf Sci. 97, 1–9. doi:10.1016/j.
Alevizos, E., Snellen, M., Simons, D., Siemes, K., and Greinert, J. (2018). Multi-             ecss.2011.10.004
   angle Backscatter Classification and Sub-bottom Profiling for Improved                     Che Hasan, R., Ierodiaconou, D., Laurenson, L., and Schimel, A. (2014). Integrating
   Seafloor Characterization. Mar. Geophys. Res. 39, 289–306. doi:10.1007/                      Multibeam Backscatter Angular Response, Mosaic and Bathymetry Data for
   s11001-017-9325-4                                                                           Benthic Habitat Mapping. PLoS ONE 9 (5), e9739. doi:10.1371/journal.pone.
Amiri-Simkooei, A., Snellen, M., and Simons, D. G. (2009). Riverbed Sediment                   0097339
   Classification Using Multi-Beam echo-sounder Backscatter Data. J. Acoust. Soc.            Cogan, C. B., Todd, B. J., Lawton, P., and Noji, T. T. (2009). The Role of marine
   Am. 126, 1724. doi:10.1121/1.3205397                                                        Habitat Mapping in Ecosystem-Based Management. ICES J. Mar. Sci. 66 (9),
Ankerst, M., Breunig, M. M., Kriegel, H.-P., and Sander, J. (1999). Optics. SIGMOD             2033–2042. doi:10.1093/icesjms/fsp214
   Rec. 28 (2), 49–60. doi:10.1145/304181.304187                                            Comaniciu, D., and Meer, P. (2002). Mean Shift: A Robust Approach toward
Applied Physics Laboratory -, A. P. L. (1994). APL-UW High-Frequency Ocean                     Feature Space Analysis. IEEE Trans. Pattern Anal. Machine Intell. 24 (5),
   Environmental Acoustic Models Handbook. Seattle, WA, USA: University of                     603–619. doi:10.1109/34.1000236
   WashingtonApplied Physics Laboratory, University of Washington. Technical                Costa, B. (2019). Multispectral Acoustic Backscatter: How Useful Is it for Marine
   Report APL-UW TR9407.                                                                       Habitat Mapping and Management? J. Coastal Res. 35 (5), 1062. doi:10.2112/
Augustin, J. M., Le Suave, R., Lurton, X., Voisset, M., Dugelay, S., and Satra, C.             jcoastres-d-18-00103.1
   (1996). Contribution of the Multibeam Acoustic Imagery to the Exploration of             Diesing, M., Mitchell, P. J., O’Keeffe, E., Gavazzi, G. O. A. M., and Bas, T. L. (2020).
   the Sea-Bottom. Mar. Geophys. Researches 18, 459–486. doi:10.1007/                          Limitations of Predicting Substrate Classes on a Sedimentary Complex but
   BF00286090                                                                                  Morphologically Simple Seabed. Remote Sensing 12 (20), 3398. doi:10.3390/
Bastos, A. C., Quaresma, V. S., Marangoni, M. B., D’Agostini, D. P., Bourguignon,              rs12203398
   S. N., Cetto, P. H., et al. (2015). Shelf Morphology as an Indicator of                  Diesing, M., Mitchell, P., and Stephens, D. (2016). Image-based Seabed
   Sedimentary Regimes: A Synthesis from a Mixed Siliciclastic-Carbonate                       Classification: What Can We Learn from Terrestrial Remote Sensing? ICES
   Shelf on the Eastern Brazilian Margin. J. South Am. Earth Sci. 63, 125–136.                 J. Mar. Sci. 73 (10), 2425–2441. doi:10.1093/icesjms/fsw118
   doi:10.1016/j.jsames.2015.07.003                                                         Dominguez, J. M. L. (2006). The Coastal Zone of Brazil: An Overview. J. Coastal
Blondel, P., and Gómez Sichi, O. (2009). Textural Analyses of Multibeam Sonar                  Res., 16–20.
   Imagery from Stanton Banks, Northern Ireland continental Shelf. Appl. Acoust.            Ester, M., Kriegel, H., Sander, J., and Xu, X. (1996). “A Density-Based Algorithm
   70, 1288–1297. doi:10.1016/j.apacoust.2008.07.015                                           for Discovering Clusters in Large Spatial Databases with Noise,” in
Brooke, B. P., Nichol, S. L., Huang, Z., and Beaman, R. J. (2017). Palaeoshorelines            Proceedings of 2nd International Conference on Knowledge Discovery
   on the Australian continental Shelf: Morphology, Sea-Level Relationship and                 and Data Mining (Munich: Institute for Computer Science, University of
   Applications to Environmental Management and Archaeology. Continental                       Munich), 226–231. (KDD-96).
   Shelf Res. 134, 26–38. doi:10.1016/j.csr.2016.12.012                                     Fakiris, E., Blondel, P., Papatheodorou, G., Christodoulou, D., Dimas, X., Georgiou,
Brown, C., Beaudoin, J., Brissette, M., and Gazzola, V. (2019). Multispectral                  N., et al. (2019). Multi-Frequency, Multi-Sonar Mapping of Shallow Habitats-
   Multibeam Echo Sounder Backscatter as a Tool for Improved Seafloor                           Efficacy and Management Implications in the National Marine Park of
   Characterization. Geosciences 9 (3), 126. doi:10.3390/geosciences9030126                    Zakynthos, Greece. Remote Sensing 11, 461. doi:10.3390/rs11040461
Brown, C. J., Smith, S. J., Lawton, P., and Anderson, J. T. (2011). Benthic Habitat         Feldens, P., Schulze, I., Papenmeier, S., Schönke, M., and Schneider von Deimling,
   Mapping: A Review of Progress towards Improved Understanding of the Spatial                 J. (2018). Improved Interpretation of Marine Sedimentary Environments Using
   Ecology of the Seafloor Using Acoustic Techniques. Estuarine, Coastal Shelf Sci.             Multi-Frequency Multibeam Backscatter Data. Geosciences 8 (6), 214. doi:10.
   92 (3), 502–520. doi:10.1016/j.ecss.2011.02.007                                             3390/geosciences8060214
Buscombe, D., and Grams, P. (2018). Probabilistic Substrate Classification with              Fezzani, R., and Berger, L. (2018). Analysis of Calibrated Seafloor Backscatter for
   Multispectral Acoustic Backscatter: A Comparison of Discriminative and                      Habitat Classification Methodology and Case Study of 158 Spots in the Bay of
   Generative Models. Geosciences 8 (11), 395. doi:10.3390/geosciences8110395                  Biscay and Celtic Sea. Mar. Geophys. Res. 39 (1–2), 169–181. doi:10.1007/
Campello, R. J. G. B., Moulavi, D., and Sander, J. (2013). “Density-Based Clustering           s11001-018-9342-y
   Based on Hierarchical Density Estimates,” in Advances in Knowledge Discovery             Fonseca, L., Brown, C., Calder, B., Mayer, L., and Rzhanov, Y. (2009). Angular Range
   and Data Mining. Editors J. Pei, V. S. Tseng, L. Cao, H. Motoda, and G. Xu,                 Analysis of Acoustic Themes from Stanton Banks Ireland: A Link between Visual
   (Springer Berlin Heidelberg), 7819, 160–172. doi:10.1007/978-3-642-                         Interpretation and Multibeam Echosounder Angular Signatures. Appl. Acoust. 70
   37456-2_14                                                                                  (10), 1298–1304. doi:10.1016/j.apacoust.2008.09.008
Chakraborty, B., Menezes, A., Dandapath, S., Fernandes, W. A., Karisiddaiah, S.             Fonseca, L., and Mayer, L. (2007). Remote Estimation of Surficial Seafloor
   M., Haris, K., et al. (2015). Application of Hybrid Techniques (Self-Organizing             Properties through the Application Angular Range Analysis to Multibeam

Frontiers in Remote Sensing | www.frontiersin.org                                      16                                               March 2022 | Volume 3 | Article 860282
You can also read